AMD and Anthropic announced a strategic partnership on July 22 at AMD's Advancing AI event in San Francisco, agreeing to deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs inside AMD Helios rack-scale systems, with the first gigawatt coming online in the first half of 2027.
As part of the deal, AMD will make a strategic equity investment of up to $5 billion in Anthropic — turning what began as a chip-supply relationship into a direct financial stake in one of the two labs Satya Nadella complained about controlling the world's "token capital" just a day earlier.
The systems pair MI455X GPUs with AMD's newly launched 256-core EPYC "Venice" CPUs and Pensando networking, and Anthropic says it will use Claude itself to help accelerate development of AMD's open ROCm software stack — quietly extending Claude's reach into chip tooling well beyond the chatbot.
CEO Lisa Su called it "the full strength of AMD high-performance computing" meeting "Anthropic's leadership in frontier AI" — a framing that puts AMD alongside Nvidia, Google and Amazon as a genuine multi-silicon backer of Anthropic's compute roadmap.
White House Office of Science and Technology Policy Director Michael Kratsios posted on X on July 22 that "we have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model," calling it "covert industrial distillation aimed at stealing proprietary U.S. technology."
It is the first time a senior US official has publicly tied a specific Chinese model to a specific American one by name, sharpening a debate that has run since Kimi K3's July 16 release, when the 2.8-trillion-parameter open model briefly topped a major coding leaderboard and rattled American labs and investors.
Kratsios's post offered no supporting evidence, and OpenAI President Greg Brockman struck a more cautious tone the same day, telling Bloomberg that K3 is "pretty good" but that it's "too early" to conclude whether Moonshot distilled from any US model.
The accusation lands as Washington and Beijing prepare high-stakes AI talks next month, with US officials weighing further curbs on American firms' use of Chinese models — meaning this claim, evidence or not, is likely to shape the terms of that negotiation.
Alphabet reported second-quarter revenue of $119.8 billion, up 24% year-over-year, with Google Cloud accelerating to $24.77 billion and its cloud backlog reaching $514 billion, up from $462 billion in the first quarter.
The Gemini app hit 950 million monthly active users, up from 900 million in May and 750 million at the end of last year, and Google says its models now process 22 billion API tokens every minute — a scale that dwarfs most rivals regardless of who currently holds the leaderboard crown.
Capital expenditure doubled to $44.9 billion for the quarter as Google races to keep pace with demand, while YouTube advertising revenue rose 13% to $11.06 billion — a reminder that Alphabet's AI bet still rides on a very large, very profitable core business.
The numbers land the same week Gemini 3.5 Pro remains conspicuously unshipped for a third consecutive deadline, suggesting investors are, for now, rewarding Google's infrastructure and distribution over its frontier-model cadence.
Fields Medalist Terence Tao published a post on July 21 titled "A digestion of the Jacobian conjecture counterexample," walking through an explicit polynomial map from complex 3-space to itself with a constant, nonzero Jacobian determinant that is locally invertible everywhere yet not globally invertible — a long-sought counterexample bearing on one of mathematics' notable open conjectures.
What made the post travel beyond mathematics circles was a single line near the end: Tao wrote that he "used an AI chatbot to discuss various aspects of this problem and to confirm several of the calculations made here," and linked directly to the underlying conversation.
It is one of the first times a mathematician of Tao's stature has documented, in public and by name, using an AI system as a working tool for verifying a serious result — rather than merely commenting on AI's capabilities from a distance. The post climbed the front page of Hacker News within hours.
For a field that has spent two years arguing over whether language models can do real mathematics, Tao's quiet, unglamorous disclosure — cited as a checking tool, not a co-author — may be a more consequential data point than any single benchmark score.
Pastor Scott Winters filed suit against OpenAI and Sam Altman in San Francisco Superior Court on July 22, alleging negligence and the unauthorized practice of medicine after he says ChatGPT-4o repeatedly reassured him that his symptoms were "not something dangerous" and at one point suggested "recliner-based micro-recovery."
The complaint quotes the chatbot telling Winters that "God did not design your body to endlessly fail" — language the suit says delayed him from seeking emergency care. Winters was admitted to intensive care on July 13, 2025, with what doctors determined was a massive pulmonary embolism.
The lawsuit seeks damages and, notably, a court order halting ChatGPT Health pending an independent safety review — a more sweeping ask than the monetary claims alone, putting a specific consumer health feature, not just the general chatbot, in a judge's sights.
OpenAI's response was brief: "ChatGPT is not a doctor and should never be used as a substitute for medical care." The case adds to a small but growing docket testing how much responsibility AI companies bear when users treat a chatbot's reassurance as medical advice.